A Cordial Sync: Going Beyond Marginal Policies For Multi-Agent Embodied Tasks
Unnat Jain*, Luca Weihs*, Eric Kolve, Ali Farhadi, Svetlana Lazebnik, Aniruddha Kembhavi, Alexander Schwing
ECCV 2020 (spotlight)
In this work, we identify and tackle two challenges when training agents to complete tightly coordinated embodied tasks. First, existing decentralized action sampling procedures do not permit expressive joint action policies. Second, in tasks requiring close coordination, the number of failed actions dominates successful actions. To this end, we introduce SYNC-policies and CORDIAL (coordination loss) which permit expressive (i.e. beyond rank-one) joint policies for decentralized and communicative agents.